course 2 16.532.031 computational electromagnetics … › sample › pdf ›...
TRANSCRIPT
Course 2
16.532.031 Computational Electromagnetics
Prerequisites: undergraduate course on Engineering Electromagnetics or
equivalent
The required textbook is PART III of the book Electromagnetic Waves, Materials and
Computation with MATLAB, written by the instructor, to be published (August15, 2011)
by CRC Press, Taylor and Francis Group.
The text book uses the math. software MATLAB in coding the algorithms developed.
There are many examples of the MATLAB code in the text for writing the main
programs. There are five function MATLAB programs written by the instructor, whose
soft copies will be supplied by the instructor to the registrants to facilitate doing the
homework and a project. In case you are not familiar with MATLAB, the last reference is
a ‘easy to learn’, MATLAB reference. You do not have to use the MATLAB in the first 3
weeks. The homework is given so that you can do all of it it by hand computation.
Homework Problems are in the text book at the end and are listed as PX.x, X being the
chapter number and x is the serial number of the problem.. The course outline given
below lists the syllabus for the whole course. The third column gives the relevant
Sections of the hard copy for the topics listed in Column 2. The last column gives the
homework assignment.
Course Outline
16.532. Computational Electromagnetics Spring 2011
Required Text Book . Electromagnetic Waves, Materials, and Computation with
MATLAB , by Professor D. K. Kalluri published in August 2011 by CRC Press, Taylor
and Francis Group.
Wk# Date Topic Section Homework
1 01/24/2011 Formulation of E.M. Problems as Differential and Integral Equations, Finite Difference Methods, Method of Weighted Residuals
15.1 - 15.3.2
P15.2 - P15.5
2 01/31/2011 Moment Method, Finite Element Method and shape functions
15.3.3 - 15.3.5
P15.8 – P15.10
3 02/07/2011 Two Dimensional Problems: Finite Difference Method, Finite Element Method (FEM)
16.1 - 16.2
P16.1
4 02/14/2011 FEM for Poisson’s and Laplace’s Equations
16.2.1 - 16.3
P16.5
5 02/21/2011 FEM for Homogeneous Waveguide
16.4 P16.4
6 02/28/2011 First Examination
7 03/072011 FEM: second order node-based, Vector finite element techniques and edge-based shape functions
16.4.1 - 16.4.3
P16.6, P16.7
8 03/14/2011 Application of vector FEM to Homogeneous Waveguide: Characteristic impedance of Transmission Lines
16.4.4 - 16.5
P16.9,P16.10
9 03/21/2011 Moment Method, Capacitance of parallel plates taking into account fringing
16.6 P16.12
10 03/28/2011 FEM: Inhomogeneous waveguide problem; 3D-triangular elements
17.1 - 17.3.1, 17.5
Project 1
11 04/04/2011 Survey of FEM Case studies and Appendices of Chapter 16
Chapter 18, 16A - E
Project 2
12 04/11/2011 Finite Difference Time-Domain (FDTD) Method
19.1 – 19.5
P19.3, P19.4
13 04/18/2011 Continuation of FDTD and Survey of FDTD Case study
19.7 – 19.10, Chapter 20
Work on Projects
14 04/25/2011 Final Examination
15 05/02/2011 Submit Projects
Numerical Grades
There are a total of 100 points possible in the course, distributed as follows.
5 Quizes-openbook 10%
Project 1
Project 2
First Exam-openbook
Final Exam-openbook
15%
15%
30%
30%
Problems suitable for Short Quizes
P15.1, P16.2, P16.11, P16.13, P17.1
Problems suitable for Exams.
P15.6, P15.7, P16.3, P16.8, P16.14, P17.2, P19.1, P19.2
Reference Books
The text book is sufficient for the course. For completeness some references are
given below.
[1] Harrington, R. F., Field Computation by Moment Methods, Macmillan, New York,
1968.
[2] Jin Jianming, The Finite Element Method in Electromagnetics, Second Edition,
Wiley, New York, 2002.
[3] Volakis, J. L., Chatterjee, A., and Kempel, L. C., Finite Element Method for
Electromagnetics, IEEE Press, New York, 1998.
[4] Taflove, A., Computational Electrodynamics, The Finite_Difference Time_Domain,
Method, Artech House, 1995.
[5] Sadiku, N. O. M., Numerical Techniques in Electromagnetics, CRC Press, 1992.
[6] Pratap, R., Getting Started with MATLAB 5, A Quick Introduction for Scientists and
Engineers, Oxford University Press, Oxford, 1999.
Undergraduate Books
Applied Electromagnetics by Ulaby
Electromagnetic Waves by U. S. Inan & A. S. Inan
Electronic copy of the 5 function programs
% GLANT.m
function [S,T] = GLANT(Nn,Ne,n1L,n2L,n3L,xn,yn);
% Global Assembly two dimensional node based
% triangular elements
for e = 1:Ne;
n(1,e) = n1L(e);
n(2,e) = n2L(e);
n(3,e) = n3L(e);
end
% Initialization
S = zeros(Nn,Nn);
T = zeros(Nn,Nn);
% Loop through all elements
for e = 1: Ne;
% coordinates of the element nodes
for i = 1:3;
x(i) = xn(n(i,e));
y(i) = yn(n(i,e));
end
% compute the element matrix entries
b(1) = y(2) - y(3);
b(2) = y(3) - y(1);
b(3) = y(1) - y(2);
c(1) = x(3) - x(2);
c(2) = x(1) - x(3);
c(3) = x(2) - x(1);
Area = 0.5 * abs ( b(2)*c(3) - b(3) *c(2));
% Compute the elemnt matrix entries
for i = 1:3;
for j = 1:3;
Se(i,j) = (0.25/Area)*(b(i)*b(j) + c(i) *c(j));
if i ==j
Te(i,j) = Area/6;
else
Te(i,j) = Area/12;
end
% Assemble the Element matrices into Global FEM System
S(n(i,e),n(j,e)) = S(n(i,e),n(j,e))+ Se(i,j);
T(n(i,e),n(j,e)) = T(n(i,e),n(j,e))+ Te(i,j);
end
end
end
% PGLANT2.m
function [S,T,g] =
PGLANT2(Nn,Ne,n1L,n2L,n3L,Rho,Epr,xn,yn);
% Solution of Poisson's Equation
% Laplacian of V = -Rho/(epsilon0*epr)
% Rho is the array of values of volume charge density in
each element
% Epr is the array of values of dielectric constant in each
element
% Global Assembly two dimensional node based
% triangular elements
for e = 1:Ne;
n(1,e) = n1L(e);
n(2,e) = n2L(e);
n(3,e) = n3L(e);
end
% Initialization
S = zeros(Nn,Nn);
T = zeros(Nn,Nn);
g = zeros(Nn);
% Loop through all elements
for e = 1: Ne;
% coordinates of the element nodes
for i = 1:3;
x(i) = xn(n(i,e));
y(i) = yn(n(i,e));
end
% compute the element matrix entries
b(1) = y(2) - y(3);
b(2) = y(3) - y(1);
b(3) = y(1) - y(2);
c(1) = x(3) - x(2);
c(2) = x(1) - x(3);
c(3) = x(2) - x(1);
Area = 0.5 * abs ( b(2)*c(3) - b(3) *c(2));
% Compute the elemnt matrix entries
ge = Rho(e)*Area/3;
for i = 1:3;
g(n(i,e)) = g(n(i,e)) + ge;
for j = 1:3;
Se(i,j) = (0.25/Area)*(b(i)*b(j) + c(i)
*c(j))*Epr(e)*8.854*10^(-12);
if i ==j
Te(i,j) = Area/6;
else
Te(i,j) = Area/12;
end
% Assemble the Element matrices into Global FEM System
S(n(i,e),n(j,e)) = S(n(i,e),n(j,e))+ Se(i,j);
T(n(i,e),n(j,e)) = T(n(i,e),n(j,e))+ Te(i,j);
end
end
end
% GLANT2T.m
function [S,T] =
GLAN2T(Nn,Ne,n1L,n2L,n3L,n4L,n5L,n6L,xn,yn);
% Global Assembly two dimensional node based
% triangular elements of second order
% written by D. K. Kalluri
for e = 1:Ne;
n(1,e) = n1L(e);
n(2,e) = n2L(e);
n(3,e) = n3L(e);
n(4,e) = n4L(e);
n(5,e) = n5L(e);
n(6,e) = n6L(e);
end
Q1 = (1/6)*[0,0,0,0,0,0;0,8,-8,0,0,0;0,-8,8,0,0,0;
0,0,0,3,-4,1;0,0,0,-4,8,-4;0,0,0,1,-4,3];
Q2 = (1/6)*[3,0,-4,0,0,1;0,8,0,0,-8,0;-4,0,8,0,0,-4;
0,0,0,0,0,0;0,-8,0,0,8,0;1,0,-4,0,0,3];
Q3 = (1/6)*[3,-4,0,1,0,0;-4,8,0,-4,0,0;0,0,8,0,-8,0;
1,-4,0,3,0,0;0,0,-8,0,8,0;0,0,0,0,0,0];
Tek = (1/180)*[6,0,0,-1,-4,-1;0,32,16,0,16,-4;0,16,32,-
4,16,0;
-1,0,-4,6,0,-1;-4,16,16,0,32,0;-1,-4,0,-1,0,6];
% Initialization
S = zeros(Nn,Nn);
T = zeros(Nn,Nn);
% Loop through all elements
for e = 1: Ne;
% coordinates of the element nodes
x(1) = xn(n(1,e));
x(2) = xn(n(4,e));
x(3) = xn(n(6,e));
y(1) = yn(n(1,e));
y(2) = yn(n(4,e));
y(3) = yn(n(6,e));
% compute the element matrix entries
b(1) = y(2) - y(3);
b(2) = y(3) - y(1);
b(3) = y(1) - y(2);
c(1) = x(3) - x(2);
c(2) = x(1) - x(3);
c(3) = x(2) - x(1);
Area = 0.5 * abs ( b(2)*c(3) - b(3) *c(2));
COTTH1 = -(0.5/Area)*(b(2)*b(3) + c(2)*c(3));
COTTH2 = -(0.5/Area)*(b(3)*b(1) + c(3)*c(1));
COTTH3 = -(0.5/Area)*(b(1)*b(2) + c(1)*c(2));
Te = Area*Tek;
Se = COTTH1*Q1 + COTTH2*Q2 + COTTH3*Q3;
% Compute the elemnt matrix entries
for i = 1:6;
for j = 1:6;
% Assemble the Element matrices into Global FEM
System
S(n(i,e),n(j,e)) = S(n(i,e),n(j,e))+ Se(i,j);
T(n(i,e),n(j,e)) = T(n(i,e),n(j,e))+ Te(i,j);
end
end
end
% GLAET.M
function [E,F] = glaet(Neg, Nn, Ne, n1L, n2L, n3L, n1EL,
n2EL, n3EL, xn, yn)
for e = 1:Ne
n(1,e) = n1L(e);
n(2,e) = n2L(e);
n(3,e) = n3L(e);
ne(1,e) = n1EL(e);
ne(2,e) = n2EL(e);
ne(3,e) = n3EL(e);
end
E=zeros(Neg, Neg);
F=zeros(Neg, Neg);
for e = 1:Ne
for i = 1:3;
x(i) = xn(n(i,e));
y(i) = yn(n(i,e));
end
b(1)=y(2)-y(3);
b(2)=y(3)-y(1);
b(3)=y(1)-y(2);
c(1)=x(3)-x(2);
c(2)=x(1)-x(3);
c(3)=x(2)-x(1);
Area=0.5*abs(b(2)*c(3)-b(3)*c(2));
l(1)=sqrt(b(3)*b(3)+c(3)*c(3));
l(2)=sqrt(b(1)*b(1)+c(1)*c(1));
l(3)=sqrt(b(2)*b(2)+c(2)*c(2));
for i=1:3
for j = 1:3
ff(i,j)=b(i)*b(j)+c(i)*c(j);
end
end
G(1,1)=2*(ff(2,2)-ff(1,2)+ff(1,1));
G(2,2)=2*(ff(3,3)-ff(2,3)+ff(2,2));
G(3,3)=2*(ff(1,1)-ff(3,1)+ff(3,3));
G(2,1)=G(1,2);
G(1,3)=ff(2,1)-2*ff(2,3)-ff(1,1)+ff(1,3);
G(3,1)=G(1,3);
G(2,3)=ff(3,1)-ff(3,3)-2*ff(2,1)+ff(2,3);
G(3,2)=G(2,3);
for i=1:3
for j=1:3
Ee(i,j)=(1/Area)*(l(i)*l(j));
Fe(i,j)=(1/48)*(1/Area)*l(i)*l(j)*G(i,j);
if(ne(i,e)<0)
Ee(i,j)=-Ee(i,j);
Fe(i,j)=-Fe(i,j);
else;
end;
if(ne(j,e)<0);
Ee(i,j)=-Ee(i,j);
Fe(i,j)=-Fe(i,j);
else;
end;
ane(i,e)=abs(ne(i,e));
ane(j,e)=abs(ne(j,e));
E(ane(i,e),ane(j,e))=E(ane(i,e),ane(j,e))+Ee(i,j);
F(ane(i,e),ane(j,e))=F(ane(i,e),ane(j,e))+Fe(i,j);
end
end
end
% INHWGD.m function
[Att,Btt,Btz,Bzz,C]=INHWGD(epr,mur,k0,Neg,Nn,Ne,...
n1L,n2L,n3L,n1EL,n2EL,n3EL,xn,yn);
% Inhomogeneous waveguide problem, page 73-75 text
% written by D. K. Kalluri
% Edge elements for transverse fields and node based
% elements for longitudinal fields
% Notation of text modified to remove confusion
% Text( Equation 3.39 ) My notation
% [St] Att
% [Tt] Btt
% [Sz] Bzz
% [G] Btz
% Global Assembly of two dimensional edge based
% triangular elements
% Neg= number of edge elements
% Nn = number of nodes
% Ne = number of elements
% n1L = array containing global node number
% of the first local node of e th elemnt
% n2L = array containing global node number
% of the second local node of e th elemnt
% % n3L = array containing global node number
% of the third local node of e th elemnt
% n1EL = array containing global edge number
% of the first local edge of e th elemnt
% n2EL = array containing global edge number
% of the first local edge of e th elemnt
% n3EL = array containing global edge number
% of the first local edge of e th elemnt
% a negative value indicates opposite directions
% of the global edge and the local edge
for e = 1:Ne;
n(1,e) = n1L(e);
n(2,e) = n2L(e);
n(3,e) = n3L(e);
ne(1,e) = n1EL(e);
ne(2,e) = n2EL(e);
ne(3,e) = n3EL(e);
end
% Initialization
Att = zeros(Neg,Neg);
Btt = zeros(Neg,Neg);
Btz = zeros(Neg,Nn);
Bzz = zeros(Nn,Nn);
% Loop through all elements
for e = 1: Ne;
% coordinates of the element nodes
for i = 1:3;
x(i) = xn(n(i,e));
y(i) = yn(n(i,e));
end
% compute the element matrix entries
b(1) = y(2) - y(3);
b(2) = y(3) - y(1);
b(3) = y(1) - y(2);
c(1) = x(3) - x(2);
c(2) = x(1) - x(3);
c(3) = x(2) - x(1);
Area = 0.5 * abs ( b(2)*c(3) - b(3) *c(2));
l(1) = sqrt(b(3)*b(3) + c(3)*c(3));
l(2) = sqrt(b(1)*b(1) + c(1)*c(1));
l(3) = sqrt(b(2)*b(2) + c(2)*c(2));
K(1) = l(1)/(12*Area);
K(2) = l(2)/(12*Area);
K(3) = l(3)/(12*Area);
% Compute the elemnt matrix entries
for i = 1:3;
for j = 1:3;
ff(i,j) = b(i)*b(j) + c(i)*c(j);
Se(i,j) = (0.25/Area)*(b(i)*b(j) + c(i) *c(j));
if i ==j
Te(i,j) =Area/6;
else
Te(i,j) =Area/12;
end
end;
end;
G(1,1) = 2*(ff(2,2) - ff(1,2)+ ff(1,1));
G(2,2) = 2*(ff(3,3) - ff(2,3)+ ff(2,2));
G(3,3) = 2*(ff(1,1) - ff(3,1) + ff(3,3));
G(1,2) = ff(2,3) - ff(2,2) - 2*ff(1,3) +ff(1,2);
G(2,1) = G(1,2);
G(1,3) = ff(2,1) - 2*ff(2,3)- ff(1,1) +ff(1,3);
G(3,1) = G(1,3);
G(2,3) = ff(3,1) - ff(3,3) - 2*ff(2,1)+ ff(2,3);
G(3,2) = G(2,3);
Gtz(1,1) = -K(1)*(b(1)*(b(1)-b(2))+c(1)*(c(1) -c(2)));
Gtz(2,2) = -K(2)*(b(2)*(b(2)-b(3))+c(2)*(c(2) -c(3)));
Gtz(3,3) = -K(3)*(b(3)*(b(3)-b(1))+c(3)*(c(3) -c(1)));
Gtz(1,2) =K(1)*(b(2)*(b(2)-b(1))+c(2)*(c(2) -c(1)));
Gtz(2,3) =K(2)*(b(3)*(b(3)-b(2))+c(3)*(c(3) -c(2)));
Gtz(3,1) =K(3)*(b(1)*(b(1)-b(3))+c(1)*(c(1) -c(3)));
Gtz(1,3) = -K(1)*(-y(1)*b(2)-y(2)*b(1)-y(3)*b(3)...
+x(1)*c(2)+x(2)*c(1)+x(3)*c(3));
Gtz(2,1) = -K(2)*(-y(2)*b(3)-y(3)*b(2)-y(1)*b(1)...
+x(2)*c(3)+x(3)*c(2)+x(1)*c(1));
Gtz(3,2) = -K(3)*(-y(3)*b(1)-y(1)*b(3)-y(2)*b(2)...
+x(3)*c(1)+x(1)*c(3)+x(2)*c(2));
for i = 1:3;
for j = 1:3;
Ee(i,j) = (1/Area)*(l(i)*l(j));
Fe(i,j) = (1/48)*(1/Area)*l(i)*l(j)*G(i,j);
% Assemble the Element matrices into Global FEM
System
if (ne(i,e) < 0);
Ee(i,j) = - Ee(i,j);
Fe(i,j) = -Fe(i,j);
Gtz(i,j) = -Gtz(i,j);
else;
end;
if (ne(j,e) < 0);
Ee(i,j) = - Ee(i,j);
Fe(i,j) = -Fe(i,j);
else;
end;
Atte(i,j) = (1/mur(e))*Ee(i,j) -
k0*k0*epr(e)*Fe(i,j);
Btte(i,j) = (1/mur(e))*Fe(i,j);
Btze(i,j) = (1/mur(e))*Gtz(i,j);
Bzze(i,j) = (1/mur(e))*Se(i,j) -
k0*k0*epr(e)*Te(i,j);
% for storing purposes take absolute value of
% ne(i,e) and ne(j,e) wich may have negative
values
ane(i,e) = abs(ne(i,e));
ane(j,e) = abs(ne(j,e));
Att(ane(i,e),ane(j,e)) = Att(ane(i,e),ane(j,e))+
Atte(i,j);
Btt(ane(i,e),ane(j,e)) = Btt(ane(i,e),ane(j,e))+
Btte(i,j);
Btz(ane(i,e),n(j,e)) = Btz(ane(i,e),n(j,e))+
Btze(i,j);
Bzz(n(i,e),n(j,e)) = Bzz(n(i,e),n(j,e))+
Bzze(i,j);
end
end
end
C1 = Btt - Btz*inv(Bzz)*Btz';
C2 = inv(C1);
C = C2*Att;
Courtesy of CRC Press/Taylor & Francis Group
y
b
ax
PEC, Ez = 0, that is, F = 0
ε, μ
2 2
n
n
∇t F + kc F = 0
Figure 3.1
003x001.eps
Courtesy of CRC Press/Taylor & Francis Group
y
b
ax
PEC, = 0, that is, = 0
ε, μn
n
∂Hz∂n
∂G∂n
2 2∇t G + Kc
G = 0
~
Figure 3.2
003x002.eps
Courtesy of CRC Press/Taylor & Francis Group
0 1 2 5
TE10
TE01
TE20
TE11
TM11
3 fc/( fc)TE10
Figure 3.3
003x003.eps
Courtesy of CRC Press/Taylor & Francis Group
y
xε, μ
a
PEC
Figure 3.4
003x004.eps
Courtesy of CRC Press/Taylor & Francis Group
y
xε, μ
a
PEC
α
Figure 3.5
003x005.eps
Courtesy of CRC Press/Taylor & Francis Group
a
a
ε1
ε1
ε2
ε2
ε
ρ
Figure 3.6
003x006.eps
Courtesy of CRC Press/Taylor & Francis Group
1.6
1.5
1.4
1.3
1.2
1.1
1.00.2 0.4 0.6 0.8 1.0 1.2 1.3
0.613
HE11
2a/λ0
β z/β
0
TE01TM01
Figure 3.7
003x007.eps
Computational ElectromagneticsWeek 2
Instructor:
1
What will be covered today
• Moment Method
• Finite Element Method and Shape Function
2
Moment Method
. . . . . . (15.65)
. . . (15.66)
3
Moment method is one of weighted residual method discussed in the previous class.Let the problem be stated as:
where L is an operator, f is the unknown function and g is excitation. The problem becomes; to find f given L and g and the boundary conditions.
Assume
gLf
N
nnn ff
1
where fn are the basic functions chosen for the problem.Choose the weight functions wm for the problem to minimize the residual. Choose the inner product for the problem.
Let us illustrate the process by giving a simple example.
Moment Method
. . . . . . (15.67a)
. . . . . . (15.67b)
4
Solve the following problem by moment method.
with the domain given as 0 < x < 1 and the boundary conditions are given by
By moment method, substitute (15.66) in (15.65), multiply by wm and form the inner product.
2
2
2
41 xdx
fd
f 0 f 1 0
The exact answer to the problem is
326
5 42 xxxxf
gwfLw m
N
nnnm ,,
1
where is the symbol for the inner product.
. . . . . . (15.67c)
. . . . . . (15.68)
Moment Method
. . . . . . (15.69)
. . . . . . (15.70)
5
For the problem, the inner product of 2 functions can be expressed as
Using (15.69) in (15.68) one obtains
(1) Pulse function P(x)
Now we need to choose the wm and fn for the problem. This needs some level of skills.Here we define two sub-sectional basic functions.
Figure 15.7 sketches this function.
1
m n m n
0
, dx
dxgwdxfLw m
N
nnnm
1
0
1
0 1
otherwise
NxxPLet
0
12
11
. . . (15.71)
xP
x
1
1
1
N
Moment Method
. . . . . . (15.72)
. . . . . . (15.73)
6
The center of the pulse function can be shifted to x = xn by defining
(2) Triangle function T(x-xn)
Figure 15.8 sketches this function.
otherwise
NxxxxP nn
0
12
11
1
10
1
111
Nxx
NxxNxxxxT
n
nnn
nxxT
1
1
2
N
xnx
Moment Method
. . . . . . (15.74)
. . . . . . (15.75)
7
After choosing wm and fn in (15.70), the result may be written as an algebraic equation
where
Equation (15.74) yields N equations for the N unknowns n.
m
N
nnmn g
1
dxLfw nmmn 1
0
1
m m
0
g w gdx . . . . . . (15.76)
We know that integration, when an impulse function is in the integral, is given by
b
j j j j
a
x x f x dx f x f a x b
0 Otherwise
. . . . . . (15.77)
Thus we can get away by choosing fn such that L fn is a sum of impulses.
Moment Method
8
This will be the result if fn =T(x-xn) and L is a second order derivative shown in Figure 15.9
Let us illustrate the calculations by choosing N =2 in (15.66). The weight functions w1 and w2, the basis functions T 1and T2 are sketched in Figure 15.10
11, f
22 , f
x
x
1
1
6
1
6
2
6
3
6
4
6
5
6
6
2
2
____ f
1
1
____ f
Moment Method
9
1 and 2 can be obtained by two algebraic equations.
The first equation is
. . . . . . (15.78)
6
4
6
223
2
12
1 xxxdx
fdLf
66
4
6
223
6/3
6/1
1
0
1111
dxxxxdxLfw
36
6
6
42
6
23
6/3
6/1
12
dxxxx
81
4041
6/3
6/1
21
0
11 dxxdxgwg
1212111 g
81
4036 21
Moment Method
10
Solving 1 and 2, we get
. . . . . . (15.80)
Similarly, we can get the second equation
81
7663 21 . . . . . . (15.79)
2222121 g
243
64,
243
5221
So the approximate solution for moment method is
6
4
243
64
6
2
243
5221 xTxTf
. . . . . . (15.81)
The accuracy can be further improved by choosing a larger value for N but the number of equations will increase as well.
Finite Element Method
11
Two aspects to the finite element method
(ii) The second aspect of the finite element method is in the technique of generating the algebraic equations.
(i) A continuous domain is broken up into a finite number of elements. Figure 15.11 shows an example. The discrete points are the vertices of the triangle.
e
The method aims to find the unknown potentials at the finite number of discrete points.
Finite Element Method – Variational Principle
12
provided the operator L is positive definite, which can be satisfied if
. . . . . . (15.82)
Instead of solving the equilibrium equation (15.65) directly, we try to find the function fthat extremizes its ‘functional’ I(f). For a functional the argument itself is a function. The functional for (15.65), in the language of linear spaces is given by
In the above X, X1, X2 are arbitrary functions that satisfy the same boundary conditions. Refer to table 15.3 for the common partial differential equations of electromagnetics and their functionals.
I f Lf , f 2 f ,g
1 2 1 2LX , X X ,LX
LX , X 0, for any X
. . . . . . (15.83)
Finite Element Method – Variational Principle
13
Let us use an example to illustrate application of (15.82). Figure 15.12 gives a one dimensional example.
with the boundary conditions
0 V
xv
0 V
x
d
md 1
The plate dimensions are assumed to be large compared to d. Poission’s equation, which can be simplifies to the one dimensional equation
v
dx
d
2
2. . . . . . (15.85)
010
Defining the inner product for the problem as
dx1
0
2121 ,
. . . . . . (15.86)
. . . . . . (15.87)
Finite Element Method – Variational Principle
14
The functional for (15.85) can be written as
The function (x) that minimized (15.89) is the solution of (15.85).
Integrating the first integral on the RHS of (15.88) and using the boundary conditions (15.86), we get
. . . . . . (15.89)
Notice that the functional (15.90) is the electric potential energy.
. . . . . . (15.90)
dx
xxdx
dx
dI v
1
0
1
0
2
2
2 . . . . . . (15.88)
21 1v
0 0
xdI dx 2 x dx
dx
21 1
E v
0 0
1 1 dI I dx x x dx
2 2 dx
Finite Element Method – Variational Principle
15
The potential function (x) in each element can be expressed by using linear interpolation:
Figure 15.13 Element (e) with end points i and j. . . (15.91)
Figure 15.14 Linear interpolation
. . . (15.92)
i j e
j i
i i
j i
x x xx x
x
xix
jx
x
x
xj
xi
Another way of writing (15.91) is
jjii xNxNx
where
ij
j
ixx
xxN
ij
ij
xx
xxN
. . . . . . (15.93a)
. . . . . . (15.93b)
Ni(x) and Nj(x) are called shape functions.
Finite Element Method – Variational Principle
16
From (15.92) and (15.93)
. . . . . . (15.98)
. . . . . . (15.99)
From (15.90)
which can be further simplified
. . . . . . (15.100)
. . . . . . (15.101)
The total functional can be obtained by summing up the functional for all the elements
ij
ij
j
ij
i
ij xxxxxxdx
d
11
11
1
dxxxNxNdx
xxI v
x
x
jjiiij
ij
x
x
ej
i
j
i
2
2
1
2
1
dxxxNdxxxN
xxI v
x
x
jj
x
x
vii
ij
ijej
i
j
i
2
2
1
elements
eE II
The algebraic equations are then obtained by minimizing (15.101) with respect to each of the unknown potentials:
k
k
EI
,0 (unknown node potential). . . . . . . (15.102)
Finite Element Method – Solve the problem
17
Assume d = 1, = 1, = 1 in Figure 15.11 and let the domain be divided into 2 elements as shown in Figure 15.16
For element (1) 1 x 2
0 1
1 2 3
2
1,0 ji xx
102 1 2 , 2
1 10 0
2 2
ji
xx
N x x N x x
i j1 20;
2 1/ 2 1./ 2
1 2
20 0
1/ 2222 2
2 2 2
0
1I 0 1 2x dx 2xdx
12 02
x0 2
2 4
Finite Element Method – Solve the problem
18
For element (2)
1,2
1 ji xx
12
2
11
2
1
,12
2
11
1
x
xxNx
xxN ii
i j2 3; 0
21 1
22
21/ 2 1/ 2
12
22 2
2 2 21
2
01I 2 1 x dx 0 2x 1 dx
12 12
x2 x 0
2 4
1 2 22E 2
I I I 22
Finite Element Method – Solve the problem
19
For minimum IE
E
2
2
I 10, 4 0
2
2
1
8
The exact answer for the problem is:
xx 12
1
8
1
2
1
2
1
2
1
2
1
The exact answer given in (15.105) coincides with the approximate answer by finite element method in (15.103); however it may be noted that this is not true in the entire domain. For example, (1/4) by exact answer (15.105) is
. . . . . . (15.105)
. . . . . . (15.106)
32
3
4
11
4
1
2
1
4
1
. . . . . . (15.107)
Finite Element Method – Solve the problem
20
By finite element method we note that x=1/4 is not an end point of an elementbut it is in the domain of element(1). In this domain
The source of the error is obvious, the exact solution shows that the potential varies quadratically where as the finite element method we used assumed a linear interpolation. The shape functions are obtained based on (15.94) and are called First order shape functions. One can define second order shape functions based on quadratic interpolation.
. . . . . . (15.108)
xxNxxN ji 2,21
i i j j
1 2
x N x N x
1 2x 2x
32
2
8
1
4
120
4
21
4
1
21
1. No questions for the week 2
2. Home work: P15.6 – P15.10
Questions for the week
Electomagnetic Waves and MaterialsWeek 2
Instructor:
1
What will be covered today
• Uniform plane waves
• Good conductor approximation
• Skin effect
• Boundary condition between PEC and dielectric medium
• AC resistance of round wires
• Bounded transmission line
2
Uniform plane wavesWhat is a sourceless medium:
v = 0 and
Jsource = 0
. . . . . . (2.1)
. . . (2.2)
. . . . . . (2.3)
. . . . . . (2.4)
3
r r One-dimensional solution for a simple lossy medium with parametersin Cartesian coordinates :
0 zz HE
EH z
jkz
jkz
te
eEE
where
2 2k j
is complex.
Uniform plane wavesThe characteristic impedance is also complex
. . . . . . (2.5)
. . . (2.6)
. . . . . . (2.7)
. . . . . . (2.8)
4
Define:
where loss tangent T is defined as
2/1
j
j
k j
where, is the attenuation constant (Np/m) is the phase constant (rad/m).
Which can be obtained by solving the equation (2.4),
2/1
2 112
1
T
2/1
2 112
1
T
T
Uniform plane waves
For a low-loss dielectric, the loss tangent T << 1 and
. . . . . . (2.9)
. . . . . . (2.10)
. . . . . . (2.11)
5
1,/2
T
1, T
1,2
tan 1
T
Good conductor approximation
Definition for good conductor
. . . . . . (2.12)
. . . . . . (2.13)
6
1T
1,1
Tf
1,542
T
sZ
2)1( jjXRZ sss
is called the skin depth. The characteristic impedance or surface impedance
. . . . . . (2.14)
We can get
Skin effectWhen waves go through the good conductor, they will be attenuated exponentially as
. . . . . . (2.15)
. . . . . . (2.16)
7
For DC, f = 0, current density is uniform inside the good conductor
. . . . . . (2.17)
where
)/cos(),( /0 zteEtzE z
)/cos(),( /0 zteJtzJ z
00 EJ
for the time-harmonic AC case (f 0), we define skin depth
f/1
• The amplitude of the current density drops to e-1 (38.8%) for a distance of .
• It will drop to zero (practically) for a distance of 4.
• At high frequencies the current is practically confined to the skin of the conductor
and the phenomenon is hence described as skin effect.
B.C. between PEC and dielectric mediumFor a perfect conductor, = => = 0, we can get E = 0 Inside the conductor
. . . . . . (2.19)
8
. . . . . . (2.20)
If medium 1 is a perfect conductor and medium 2 is a dielectric, we can get the boundary conditions as
sn 212ˆ D
0ˆ212 Bn
0ˆ212 En
KH 212n
. . . . . . (2.21)
. . . . . . (2.22)
n is the normal unit vector on the boundary.
B.C. between PEC and dielectric mediumSince E = 0 Inside the conductor, we get: (2.24) and (2.26) see problem P2.2
. . . . . . (2.13)
9
. . . . . . (2.24)
. . . . . . (2.25)
. . . . . . (2.26)
0tE
0
n
En
0nH
0
n
H t
n is the normal direction and t is the tangential direction.
AC resistance
For an infinitely deep good conductor of conductivity defined by the half-space 0 < x < ,
shown in the figure, E and H can be expressed as in next slide:
10
ds
b
l
x
z
y
AC resistanceE and H inside half-space conductor
. . . . . . (2.27)
11
. . . . . . (2.28)
. . . . . . (2.29)
. . . . . . (2.30)
The time averaged power density
//0ˆ)(
~ jxxeEzxE
45//0
2ˆ)(
~ jjxx eeeEyx H
Real parts of the fields are
)(cosˆ),( /0
x
teEztx xE
)45(cos2
ˆ),( 0/0
x
teEytx xH
/220
/220
4
1ˆ45cos
22
1ˆ xx eExeEx S . . . . . . (2.31)
AC resistance
Total phasor current entering the conductor of width b
. . . . . . (2.33)
12
. . . . . . (2.35)
. . . . . . (2.32)
. . . . . . (2.36)
Define an AC equivalent resistance which consumes the same power as
After integrating, substituting the limits and transferring to time domain, we get
b
xj dydxeEddI0 0
/)1(0
~~~sEsJ
)45cos(2
)( 0
tbE
tI
Total power entering the conductor of width b and length l is given by
blEblPx
200 4
1S
PconsumedPowerRI ACRMS 2~
where 2
~ 0
bEI RMS is the AC RMS equivalent current.
AC resistance
When AC waves go through an infinitely deep good conductor of conductivity defined by
the half-space 0 < x < , shown in the figure 2.1, the equivalent DC conductor has d depth.
13
. . . . . . (2.38)
Finally, we get the AC equivalent resistance
b
lRAC
c
nonuniformdistribution
b
l uniformdistribution
b
l
=
AC resistance of round wires
The AC resistance of a round wire of
radius a and length l.
14
c
,
b
c
ac 2
,l l
R as a
1. >> a, we get
,2
lR a
a
2. << a, we get
. . . . . . (2.39)
. . . . . . (2.42)
Transmission linesA transmission line is modeled by distributed circuit theory,
15
where,
R’: The resistance due to imperfect conductors
L’: The inductance due to magnetic flux generated by the currents in the
conductors
G’: The conductance due to an imperfect dielectric
C’: The capacitance due to the surface charges on the conductors
V(z,t) V(z+z,t)
I(z,t) Rz Lz
CzGz
I(z+z,t)
z
Transmission linesThe parameters are computed as though the fields are static. For a coaxial cable shown (slide
14), the parameters are
16
. . . . . . (2.42)
. . . . . . (2.43)
. . . . . . (2.44)
. . . . . . (2.45)
For high frequencies, << a
)/ln(
2
abG
)/ln(2
abL
)/ln(
2
abC
baR
cc 2
1
2
1
Transmission linesL’ including the correction due to the internal inductance (inductance due to the magnetic flux in
the conductors) is given
17
. . . . . . (2.46)
. . . . . . (2.48)
. . . . . . (2.49)
where
For a two-port network shown in figure 2.4, the following equations can be obtained
LabL
)/ln(
2
RL
t
tzIzLtzzIRtzzVtzV
),(),(),(),(
t
tzILtzIR
z
tzV
),(),(
),(
when z -> 0
t
tzVCtzVG
z
tzI
),(),.(
),(
Similarly
. . . . . . (2.51)
Transmission linesFor a the lossless transmission line
18
. . . . . . (2.52)
. . . . . . (2.54)
. . . . . . (2.57)
where
If the source is harmonic
where
Solution
0,0v
12
2
22
2
GR
t
V
z
V
CL
1v
0~
~2
2
2
V
z
V
v
zjzj eVeVV 00
~~~
Transmission linesThe relation between the voltages and the current
19
. . . . . . (2.58)
. . . . . . (2.62)
where
For a non-PEC, we get
IZV~~
0
IZV~~
0
CLZ /0
. . . . . . (2.59)
CL
Parameters Comparison between TEM wave and transmission
TEM wave Transmission wave
E V
H I
C’
L’
Z0
CjG
LjRZ
0
Bounded Transmission linesFor a bounded transmission line,
20
where,
d: transmission line of length
Vg: Voltage source
Zg: internal impedance
ZL: External load
d
, Z0
ZL
Zg
A
B D
C
gV~
0
~~VVL Z(d)
z
IL
gI~
Bounded Transmission linesThe voltages and the current
21
. . . . . . (2.63)
. . . . . . (2.66)
where
Defined 0 as the reflection coefficient at the load
. . . . . . (2.64)
djdj eVeVdV 00
~~)(
~
djdj eVeVZ
dI 00
0
~~1)(
~
dj
dj
e
eZ
dI
dVdZ
20
20
01
1
)(~
)(~
)(
0
0
0
00 ~
~
ZZ
ZZ
V
V
L
L
By substituting (2.66) in (2.65), the input impedance can be obtained
djZdZ
djZdZZdZ
L
L
sin cos
sin cos)(
0
00
. . . . . . (2.65)
. . . . . . (2.67)
Bounded Transmission lines: see Apendice 2B and 2C for Smith Chart etc.
From 2.67, we get input impedance for special length
22
. . . . . . (2.68)
. . . . . . (2.71)
Two special cases
For a matched line, ZL = Z0 , 0 = 0
. . . . . . (2.69)
. . . . . . (2.70)
LZZZ /)4/( 20
LZZ )2/(
1,)(
d
ZdZ L
( ) .LZ d Z
)0(, tan )( 0 LZdjZdZ
)(,cot )( 0 LZdjZdZ
. . . . . . (2.72)
. . . . . . (2.73)
23
A lossfree nonuniform transmission line has
L’ = L’(z)C’ = C’(z)where L’ and C’ are per meter values of the series inductance and parallel capacitance of the transmission line.
Determine the partial differential equation for the instantaneous voltage V(z,t).For an exponential transmission lineL’(z ) = L0 exp(qz)C’(z) = C0 exp( - qz),
assuming V(z,t) = V0 exp [j(wt – kz)],determine the relation between w and k.
Questions for the week